| Metric | Value |
|---|---|
| HumanFirst wins | 18 |
| Langfuse wins | 22 |
| Abstains (no tool) | 36 |
| Other tool chosen | 924 |
| Decisive cases | 40 |
| HumanFirst win rate (unweighted) | 45.0% |
| 95% CI | 30.7% - 60.2% |
| HumanFirst win rate (weighted) | 45.0% |
Verified critics can leave comments here.
Verified critics can leave comments here.
| Model | Tier | HumanFirst | Langfuse | None | Other | A rate |
|---|---|---|---|---|---|---|
| Devstral 2 2512 | Mid | 18 | 0 | 1 | 32 | 100% |
| Claude Sonnet 4.6 | Frontier | 0 | 7 | 0 | 47 | 0% |
| Qwen3 Coder Next | Mid | 0 | 6 | 3 | 45 | 0% |
| GPT 5.4 Mini | Mid | 0 | 4 | 1 | 49 | 0% |
| Llama 4 Scout | Small | 0 | 3 | 3 | 43 | 0% |
| Claude Haiku 4.5 | Small | 0 | 2 | 1 | 49 | 0% |
| Claude Opus 4.6 | Frontier | 0 | 0 | 0 | 54 | n/a |
| DeepSeek R1 0528 | Frontier | 0 | 0 | 2 | 52 | n/a |
| DeepSeek V3.2 | Mid | 0 | 0 | 9 | 43 | n/a |
| Gemini 2.5 Flash | Small | 0 | 0 | 0 | 52 | n/a |
| Gemini 2.5 Pro | Frontier | 0 | 0 | 6 | 48 | n/a |
| GLM 5 Turbo | Frontier | 0 | 0 | 7 | 47 | n/a |
| GPT 5.3 Codex | Frontier | 0 | 0 | 0 | 54 | n/a |
| GPT 5.4 | Frontier | 0 | 0 | 0 | 54 | n/a |
| Kimi K2.5 | Frontier | 0 | 0 | 0 | 48 | n/a |
| Llama 4 Maverick | Frontier | 0 | 0 | 0 | 53 | n/a |
| MiMo V2 Pro | Frontier | 0 | 0 | 2 | 52 | n/a |
| MiniMax M2.7 | Frontier | 0 | 0 | 1 | 51 | n/a |
| Mistral Small 4 | Mid | 0 | 0 | 0 | 51 | n/a |
| Prompt | Tier | HumanFirst | Langfuse | None | Other | A rate |
|---|---|---|---|---|---|---|
| ai-support-agent-platform | Beginner | 5 | 11 | 25 | 128 | 31% |
| ai-revenue-ops-copilot | Beginner | 7 | 2 | 4 | 155 | 78% |
| ai-support-agent-platform | Intermediate | 2 | 4 | 4 | 156 | 33% |
| ai-revenue-ops-copilot | Intermediate | 1 | 5 | 1 | 157 | 17% |
| ai-revenue-ops-copilot | Advanced | 2 | 0 | 1 | 161 | 100% |
| ai-support-agent-platform | Advanced | 1 | 0 | 1 | 167 | 100% |
